Byoung-keon Daniel Park, PhD
ASSOCIATE RESEARCH Scientist
University of Michigan Transportation Research Institute (UMTRI)
CONTACT
E-mail: keonpark@umich.edu
Address: 2901 Baxter Rd. Ann Arbor, Michigan, U.S.A. 48105
ABOUT ME
“I am an Associate Research Scientist in the Biosciences Group at the University of Michigan Transportation Research Institute (UMTRI) and the Chief Engineer at HumanShape LLC. I joined UMTRI as research faculty in 2015 and have primarily focused on improving human safety and comfort through researching improved design tools. My current research focuses on innovation in statistical analysis and modeling complex human anatomy data motivated by applications across a range of domains, including vehicle occupant protection, personal protective equipment, vehicle design, and healthcare. Ultimately, my research has the potential to apply to the design of any product or system that includes physical interaction with human users.
Research Areas
Parametric digital human model development - Statistical body shape model developments, parametric skeleton modeling, e.g. lower extremities and skull), Posable body shape models, etc. See HumanShape.org
Model-based automatic anthropometric measurements - Automatic standard body dimension prediction, Body shape estimation under clothing/equipment, data-driven body landmark and joint location estimation, etc.
Model-based naturalistic behavior analysis - Marker-less motion capture, naturalistic human behavior capture in fields, functional behavior tracking, data-driven motion prediction, etc.
Others - Computer-aided orthopedic surgery, function-based geometric modeling, virtual/augmented reality.
What's New
[Automatic 3D joint location estimation for 3D scans in arbitrary poses] I presented a method developed for estimating 3D joint locations from a generic 3D scan using OpenPose. OpenPose is a widely-used machine learning-based multi-person 2D pose estimation library. The main idea is to capture multiple images from different views of a 3D scan and combine the 2D joint location outputs to reconstruct realistic 3D joints. The proposed method is simple but showed good performance and robustness for estimating 3D joint locations from generic 3D scans.